Analog System Fault Intelligent Diagnosis Based On Feature Space Vector

Analog circuit faults are of complexity and multiformity, fault diagnosis is too difficult.Current theories and methods on fault diagnosis can not meet the practical needs. To verylarge scale integrated (VLSI) circuits in the operation, the number of analog integratedcircuits’ faults accounts for around80%of the whole faults. If the fault elements and typescan be quickly found and predicted, the staff will timely take corresponding measures to nipfaults in the bud and enhance the safety and maintainability of devices. Therefore, carryingout analog circuit fault diagnosis research has a strong theoretical and practical significance.This paper firstly reviews the research progress of analog circuit fault diagnosis at homeand abroad, then introduces the basic theory of fault diagnosis and the usual feature extractionmethods. With respect to the current difficulty of fault feature extraction and recognition,profitting from the existing research results, puts forward a new solution in the paper.With respect to fault feature extraction, the paper applys wavelet packet decompositionand reconstruction to fault signal samples, computes their band energys, constructs initialfault feature vectors of patterns’ samples; Puts forward a novel clustering method andcalculates patterns’ fuzzy sets on each dimension within the initial feature vectors; privides asearch algorithm of candidate sets including fault optimal feature. Every feature vetor canidentify various faults, and each fault can also be identified by more than one feature vetor.With respect to fault recognition, the paper puts forward query rules of patterns’ bestneighbors, provides best neighbors query algorithms in the2-D and3-D feature space, theserules can be easily expanded to higher feature space. The target is to expand the patterns’clustering radius and to determine the largest identifying range for each pattern in the featurespace. Finally, we can choose the feature vector having the highest recognition rate as thefault optimal feature.The results of analog circuits fault diagnosis in this are better than the diagnosis resultsfrom references.These prove the effectiveness of the proposed method.